Data-Driven Approach to Assessing the Tensile Strain Capacity of Pipelines With Two Different Girth Welds

Abstract The effect of girth welds on the tensile strain capacity (TSC) of pipes is critical in the strain-based design and assessment of pipelines. In this study, machine learning (ML) models for regression and classification were developed and evaluated to predict the tensile strain capacity for typical mechanized gas metal arc welding (GMAW) and flux-cored arc welding (FCAW)/shielded metal arc welding (SMAW) pipes and to classify data from the girth-welded pipes. The regression models were trained on over 15,000 data points for each pipe, derived from TSC equations found in the literature. The classification model utilized all data points from the two types of pipes. The developed regression models demonstrated accurate predictions of the TSC for both FCAW and GMAW pipes, without overfitting or underfitting, and properly captured relationships between the TSCs and features. Random Forest’s built-in capability for computing feature importance indicated that flaw depth is a critical feature affecting the TSCs of the two girth-welded pipes. However, the performance of the classification model was unsatisfactory due to inseparable data within a certain range, although it could be improved to some extent by applying different selections of features.